Agents: Everybody’s got them. But who’s managing them? And what happens to the human talent working alongside them—and being asked to show up differently as a result? “The rise of AI doesn’t mean we all suddenly devolve our responsibilities for quality control and critical thinking,” says Kate Smaje, McKinsey’s global leader of technology and AI and coauthor of the recent book Rewired. “That’s not the world we live in. There’s a premium on those capabilities now.” In this episode of McKinsey Talks Talent, Smaje joins Senior Partner Brooke Weddle and Partner Bryan Hancock, along with Global Editorial Director Lucia Rahilly, to discuss what leading through the AI moment means for the relationship between people and technology, including ways of working; costs, benefits, and risks; and the need for open dialogue about the question, “Why AI?”
To watch the full-length video of this interview, visit McKinsey’s YouTube channel.
The following transcript has been edited for clarity and length.
Where leaders go astray in the talent calculus
Lucia Rahilly: In today’s world of head-spinning disruption, leaders are confronting a daunting set of challenges. If I’m a leader, how much do I really need to get into the nitty-gritty of upskilling myself on AI?
Kate Smaje: I’d love to be able to say, “Hey, don’t worry about it,” but that isn’t the case. You have to put on your own oxygen mask first to help your organization embrace this moment.
If you’re going to lead your organization on it, then you’ve got to lead from the front.
CEOs often spend a lot of time saying, “My organization needs to do X or Y. I need people to be more tech proficient, more fluent in what they do.” My first question is always, “What are you personally doing? What’s been your learning journey? Do you learn differently than you did before?” Some leaders have done that work and are ready with a response, which leads to an amazing conversation. But many say, “Oh, I didn’t think that was about me.” As with any topic, if you’re going to lead your organization on it, then you’ve got to lead from the front.
Lucia Rahilly: We’ve repeatedly heard that this kind of transformation is not just a tech transformation but also a business—and specifically a people—transformation. If that’s the case, why are so many leaders falling short in building the in-house talent they need to turn their tech into a meaningful business advantage? What are they doing wrong?
Kate Smaje: The technology is the easy part. Here’s the hard part: Can you create the organizational capacity and capability to absorb that level of change?
So what are leaders getting wrong? First, mindset. You can’t assume this is just the next “rack and stack” of technology spend that your organization needs to roll out. Flipping that mindset already puts you at a significant advantage.
Second, people often equate the talent transformation with “How many new people can I bring into my organization?” We’ve all done it. You celebrate the number of hires coming in and the places you poached them from, and that becomes the high-five moment. But have you taken the friction out of their day-to-day so they can thrive in the job, not just survive in it? If I hire a data engineer three layers below me in the org chart and I need to pay her more than my own salary, that’s a problem. It breaks the compensation model and career pathways. So rather than focusing on how much new talent you’re getting through the door, focus on how you’re setting that talent up to be wildly successful once it arrives.
Bryan Hancock: How do you think about talent within the traditional tech organization versus talent in the business that now needs to be more tech savvy and knowledgeable than ever before?
As one client put it, “It is far easier for me to teach my metallurgists AI than for me to teach my AI specialists metallurgy.
Kate Smaje: Another mistake leaders make is forgetting that core business owners and domain owners are some of their most valuable talent. These are the folks who deeply understand how the organization works and what its subject matter expertise is. They have wisdom and the learned pattern recognition you need to harness.
This isn’t all about new talent coming in, and it’s not all about technology talent. It’s about how you transform that business talent because, as one client put it, “It is far easier for me to teach my metallurgists AI than for me to teach my AI specialists metallurgy.” Every organization has its own version of those metallurgists. Figuring out how to make sure that core, intrinsic, nondelegable capability thrives is essential.
Rewiring ways of working
Brooke Weddle: How have you seen ways of working changing—for example, the rituals and routines of work? At McKinsey, for example, we’ve changed how teams operate so they’re able to embrace new profiles and AI overall.
Kate Smaje: Ultimately, it’s not about putting a technology “lick of paint” over the business you have today. It’s about reimagining a better way of doing things. Maybe that looks like imagining a process where human capacity is no longer a constraining factor, or where what used to be top quartile is now hygiene, and therefore, you can raise the ceiling again. As you start to reimagine your business, you’re forced to fundamentally change how you work.
Subscribe to the McKinsey Talks Talent podcast
Think about our team rooms today. We problem-solve with AI. The problem-solving isn’t being done by AI. One of the core mantras of our product teams is “not AI that does it for me, but AI that does it with me.” That’s why we’re spending more time on our rituals, on what we’re trying to get done, and on what capabilities our people need. Those capabilities are amplified by tools—but the point is not the tools; it’s the capabilities.
If you rethink rituals, habits, or workflows in a more corporate context, that’s how you unlock value. You’ve got to redesign those facets of your business for the role technology will play, rather than assume technology creates that change. It doesn’t.
Brooke Weddle: This is why we keep using the word “reimagine” when we talk about redefining a workflow, a domain, or even an entire organization. Some people might think that sounds exciting, but it’s really difficult. In the context of a workflow, how do you get the right cross-disciplinary and cross-functional talent in the room from the start, so you don’t “tech enable” a bad process?
Kate Smaje: Organizations aren’t as static as we’ve treated them in recent years. Even a moment of reimagination carries an implicit assumption that there’s some kind of “from–to.” The “to” part is much fuzzier and more malleable than before because these technologies will change. The bar will keep rising. What we understand about cognitive load and brain health will continue to shift. It’s a much more dynamic way of thinking about organizations.
Bryan Hancock: Several of my HR clients are thinking of the “to” as moving from a focus on centers of excellence or expertise to a focus on employee journeys. They’re loading their organizations with “people technologists” who can partner with deep technical experts and can also rewire processes on their own, using low- or no-code tools, as priorities change. Are you seeing this shift more broadly across organizations—business owners taking on more of the ongoing design once an agentic platform is in place?
Kate Smaje: Absolutely. I’ve seen this particularly in more successful businesses once they’ve integrated these tools. You’ve really hit on the point that it’s about building the organizational muscle to be able to make these changes repeatedly. I don’t like the term “AI transformation,” because it conjures up the idea that, at some point, we’ll be done and we’ll all high-five and walk off into the sunset. It just doesn’t work like that.
The fusion of people and technology is more important today than it’s ever been.
Also, the role of the CHRO [chief human resources officer] is critical. The other day, I was reflecting on how I spend my time, and I realized that I spend the most time with companies’ people teams. The fusion of people and technology is more important today than it’s ever been. Your technologists need to understand what it takes to move people and to bring in technology that changes organizational capacity. At the same time, your “people person” employees need to understand how technology changes what is realistically possible and what those changes mean for how your organization will work differently at multiple layers.
Managing a workforce of humans and agents
Bryan Hancock: To that end, a private equity CEO I work with renamed their chief people officer as the “chief performance officer.” Because AI was going to significantly change their business, their people operations were going to become more difficult to manage. However, if they got the people side right, they’d be able to improve performance in very specific, tangible ways for individual managers, teams, and units. They changed the label to really drive home the point you were just making: People are central to the new way of working.
Brooke Weddle: HR is no longer managing the performance of just humans. It now manages digital workers as well. There’s a bias today that any agent is a good thing, but that’s simply not true. Agents need to be fine-tuned over time and, in some cases, sunset, depending on business needs. Can you speak about performance management not only for humans but also for digital workers or agents?
Kate Smaje: We know how to manage the performance of our human workforce. We know how to get the best out of our people. You hire them for a certain set of skills, develop them over time, and make sure their emotional and mental well-being is strong. When the time comes to say goodbye, whether because of poor performance or retirement, you’ve got a process for that as well. But when you ask, “When did we last talk about the performance of our nonhuman labor?”—I just don’t come across many organizations that can recall prioritizing these conversations.
McKinsey Talks Talent Podcast
Bryan Hancock, Brooke Weddle, and other talent experts help you navigate a fast-changing landscape and prepare for the future of work by making talent a competitive advantage.
Your agents need good engineers to fine-tune how they work. They need access to compute because that is their energy source. They need strong life cycle management because they can’t exist in perpetuity. A wonderfully fit-for-purpose agent that existed six months ago could possibly be a significant liability today. My new favorite term is “abandonware”—the things in your networks that have been forgotten about and should no longer exist.
As you think about agent performance, ask yourself, “How many agents do I actually have? Who is responsible or accountable for their performance? What value are they creating?” For years, we’ve had conversations about “talent to value” and the ROI of human capital. Now, we need to have the same conversations about our nonhuman workforce.
Lucia Rahilly: And whom do you see taking responsibility for agents as they start to proliferate across organizations? Who’s the right person to take ownership, if there is one?
Kate Smaje: I would love to see the business owners who use these agents for their workflows take on that responsibility. I don’t think this is a technology problem. I don’t think the CTO [chief technology officer] suddenly needs to manage half the organization’s workforce.
Let’s say you’re in finance and you have a bunch of agents helping with revenue reconciliation, month-end close, and data reporting. The head of finance who’s looking after the humans in that process should also be looking after the agents, in my opinion. That means you’ll also need guardrails, standards, and clear life cycle management across your organization, to ensure a shared risk posture. I think this will be increasingly federated to where it is needed. We’ve already seen this shift occur in other forms of technology and data.
The rise of AI doesn’t mean we all suddenly delegate our responsibilities for quality control and critical thinking.
Bryan Hancock: In HR in particular, we’re seeing that some of the deep HR experts, such as the org-psych PhDs who keep us out of trouble in our assessments, need to understand agents so they can supervise them, make sure guardrails are in place, and ensure they’re operating within the lines. We need the folks who traditionally did that in a nonagentic world to be comfortable managing agents that are potentially putting the organization at risk.
Kate Smaje: The rise of AI doesn’t mean we all suddenly delegate our responsibilities for quality control and critical thinking. That’s not the world we live in. There’s a premium on those capabilities now. Your ability to parse what an agent is doing, set it up properly, understand how it thinks, and monitor the information it can access really matters.
Scaling AI without losing the plot
Bryan Hancock: What have you seen in terms of business owners collaborating with their technology teams to manage the cost of compute for agents they’ve created?
Kate Smaje: When a great idea pops up, I always ask, “What happens when it’s wildly successful?” It’s amazing how often people look at their feet and say, “I’d have to refactor the whole code base because I didn’t build it for scale,” or “I only built it as a proof of concept,” or even, “We’ll have to change how we architect the tool because it shouldn’t call on a large language model the whole time. That’s going to create huge latency problems and huge token costs.” I don’t think this problem is just within the purview of tokenomics. I don’t think there’s been enough thought given to those second- and third-order consequences of AI.
On tokenomics specifically, I see a few things. First, we’re just starting to understand, model, and forecast where these AI tools are headed. Finance and technology departments are collaborating to find ways to forecast both usage of and payment for these tools over time to achieve their desired outcomes. That’s where I see a positive reaction.
Where I see a negative reaction is when teams pull back, strangle their progress, remove their licenses, and halt their usage. That’s not a sustainable solution. The sustainable solution is to ask, “What is our appetite for this? How do we want to think about it?” You must be proactive and consider better forecasting and modeling of where this will end up.
Many folks are thinking about what this means for their vendor relationships. They’re considering how it will affect contract negotiations, as well as token marketplaces and the ability to balance high-volume users against low-volume users in their organizations. We can’t just solve for superusers, because that’s not how your whole organization uses AI. There are also considerations in the design of the software. Base models offer benefits such as being faster and cheaper than custom-built agents, particularly for basic tasks such as transcription.
Lucia Rahilly: What’s your take on how to inculcate adoption across an organization? How do we get to the right level of fluency?
Kate Smaje: The mistake is having conversations about tools. I get nervous when I see scorecards with adoption metrics such as “What’s the adoption rate of tool A versus tool B versus tool C, market by market?” That’s a highway to nowhere. It doesn’t tell you how the technology is being used, let alone how it’s creating value. What you really need to examine are your core capabilities. What are the jobs to be done? The tools are a conduit to that question, but not the answer.
Talking about AI anxiety
Brooke Weddle: You and I were with a group of CFOs yesterday, and one thing that kept coming up was that we’re living through a period of considerable fear and anxiety. We talked a lot about psychological safety. How are leading organizations creating the conditions for people to embrace AI in thoughtful ways? Are they changing how they work toward an outcome consistent with a great employee journey? Have you seen anything that’s really working?
Kate Smaje: I think there is a heightened level of anxiety, both inside and outside organizations. Inside, there’s FOBO: fear of becoming obsolete. I get a lot of feedback that people are unsure whether they can keep up with the pace of change. I totally get that.
Outside organizations, there’s a bit of an AI backlash. People think AI is bad for a variety of reasons: They don’t want a data center in their backyard, it might take jobs, or their winter fuel bill might go up due to pressure on the grid. There are all these different opinions, and I think the onus rests on leaders, corporate or otherwise, to think about how to develop positive outcomes from AI. They must ask, “How can we do something that will be net positive for our organization, net positive for those operating within it, and net positive for the communities and society in which we operate?”
I think some of these fears come from not having open dialogues. We all know that fear amplifies in a vacuum. We need to fill that vacuum. We need to have open and honest conversations about why we’re doing all this work on AI integration. This isn’t AI for AI’s sake. It’s not because we all want to be cheaper, less interesting, less culturally diverse versions of ourselves. That’s not why we’re doing it. So why are we doing it?
When I ask CEOs and other C-suite executives that “why” question, I get fascinating answers. I’ve had clients respond, “I want to be able to solve disease problems that couldn’t be solved otherwise, because I can’t speed up my R&D pipeline fast enough.” One client said, “I want to make sure that every person, wherever they are, gets the insurance backing they need so that when something goes monumentally wrong, they’re protected in a way that’s fair and just.” Another said, “I want to solve how you ‘square the circle’ of cheap, affordable, and green energy altogether.” There are many versions of these aspirational answers, but the “why” behind the plan is often super interesting.
Brooke Weddle: I always give the example of my friend who’s an ER doctor. AI is freeing her up to be a better doctor and a better human because she’s not doing all the transcribing and paperwork that medical professionals say are among their least favorite parts of the job. I think we need to lean into the human parts of what AI enables us to do.
Bryan Hancock: There’s interesting research showing that medical training done with AI creates better outcomes than traditional training. The student tries first, and then the AI offers a different suggestion. If you integrate AI thoughtfully into the apprenticeship process, you can accelerate how people learn. If you’re not thoughtful, you can end up with some of the outcomes people worry about. I’m on the optimistic end; I think AI will free people up to reach their full capabilities faster.
We need to plan for a workforce that will face a very different set of demands on its cognitive ability.
Kate Smaje: There’s a second-order effect as well. When I see the people who work most intensively with AI right now, the natural inclination is to think that it will free up a great deal of their time and be mostly wonderful. What you find is that the high-usage groups are exhausted—not because AI is failing them, but because it’s working. It’s removing all those rote tasks from their plates. What’s left is a level of cognitive engagement we’re not used to day-to-day, tasks that require real judgment—making a decision or having a difficult conversation. Those are emotionally draining.
What we’re starting to see is the cognitive load on AI superusers going up, not down. As optimistic as I am, there’s a second-order point about brain health and cognitive load that needs to be addressed. We need to plan for a workforce that will face a very different set of demands on its cognitive ability.
Humanware on the horizon
Lucia Rahilly: Are you seeing anything new emerging in your work with clients that leaders should have on their radar?
Kate Smaje: The ability to discern the difference between the last model and the next model is getting harder across many of the things we’re starting to use AI for. Part of me hopes that this feeling that the world is moving so fast, which causes a lot of fear, might start to settle a little bit. My optimistic self says we might see that feeling subside over time.
Another thing coming is that we’re going to spend more time on humanware—and maybe a little less on software and hardware. A hard part of incorporating AI is reimagining workflows. Folks need to answer the question, “How do I need to show up differently in my hybrid workforce of carbon and silicon employees to bring them together in a way that’s actually going to work?” Those questions are vexing people. If we use this moment to lean into the humanware part of the equation, I think that will help massively.
